©Copyright JASSS

Sungho Lee (2010)

Simulation of the Long-Term Effects of Decentralized and Adaptive Investmentsin Cross-Agency Interoperable and Standard IT Systems

Journal of Artificial Societies and Social Simulation 13 (2) 3<http://jasss.soc.surrey.ac.uk/13/2/3.html>

Received: 29-Dec-2008 Accepted: 15-Jul-2009 Published: 31-Mar-2010

Keywords:

1.1

Abstract

Governments have come under increasing pressure to promote horizontal flows of informationacross agencies, but investment in cross-agency interoperable and standard systems have beenminimally made since it seems to require government agencies to give up the autonomies inmanaging own systems and its outcomes may be subject to many external and interaction risks. Byproducing an agent-based model using 'Blanche' software, this study provides policy-makers with asimulation-based demonstration illustrating how government agencies can autonomously andinteractively build, standardize, and operate interoperable IT systems in a decentralized environment.This simulation designs an illustrative body of 20 federal agencies and their missions. A multiplicativeproduction function is adopted to model the interdependent effects of heterogeneous systems on jointmission capabilities, and six social network drivers (similarity, reciprocity, centrality, mission priority,interdependencies, and transitivity) are assumed to jointly determine inter-agency system utilization.This exercise simulates five policy alternatives derived from joint implementation of three policylevers (IT investment portfolio, standardization, and inter-agency operation). The simulation resultsshow that modest investments in standard systems improve interoperability remarkably, but that awide range of untargeted interoperability with lagging operational capabilities improves missioncapability less remarkably. Nonetheless, exploratory modeling against the varying parameters fortechnology, interdependency, and social capital demonstrates that the wide range of untargetedinteroperability responds better to uncertain future states and hence reduces the variances of jointmission capabilities. In sum, decentralized and adaptive investments in interoperable and standardsystems can enhance joint mission capabilities substantially and robustly without requiring radicalchanges toward centralized IT management.

Public IT Investment, Interoperability, Standardization, Social Network, Agent-Based Modeling,Exploratory Modeling

Introduction

The dominant structural forms in all the governments have been stovepipe organizational units, soinformation systems have been optimized to form agency-centric vertically integrated systems. Asmany contemporary policy challenges span multiple policy domains, public organizations have comeunder increasing pressure to promote horizontal flows of information, work and decision-makingacross functional boundaries (Agranoff 2003; Kohtamaki et al. 2008). However, studies of networkingacross jurisdictional boundaries found little to support any illusion that government could be instantly

1.2

1.3

1.4

1.5

1.6

or easily transformed (Fountain 2001; Goldsmith and Eggers 2004; Lazer and Binz-Scharf 2004).Networking governments requires two steps of transformational efforts: 1) efforts to identifycomplementary systems across agencies and improve their interoperability or reusability, and 2)inter-organizational operation of the interoperable systems to achieve missions effectively.

Information systems can become interoperable by coupling them either bilaterally (i.e. point-to-point)or multi-laterally (i.e. through a standard middleware). Compared with point-to-point couplings,integration through standard systems can achieve a wide range of interoperability. Centralizedmanagement of government-wide information systems can promote cross-agency consolidation ofpotentially redundant systems into shared services and improve the interoperability of informationsystems through standard middleware. However, centralized management is disadvantageous todecentralized management (i.e. local control of information systems) in regards to encouraging fieldoffices to understand their unique information needs and allocate scarce resources responsibly, andfacilitating user adaptation and innovation (Libicki 2000). In addition, although standardization canachieve a wide range of interoperability, the short-term return on investment may be smaller (i.e. lesscost-effective) than that of targeted point-to-point couplings that can achieve performance gainsimmediately (Kaye 2003).

After systems become made interoperable, it takes time for potential users to appreciate thecomplementary values of interoperable systems of other agencies and utilize these systems.Fountain (2002) claims that objective technology and enacted technology need to be distinguishedsince many intergovernmental information systems have not been productively utilized bygovernment workers. If investment in technological interoperability is not matched with inter-organizational operational capability, interoperable system capacities will be left under-utilized.Although system integration may be managed from a top-down perspective, inter-agency systemoperations are steered only by a variety of incentives and sanctions because they are carried out byself-governing agencies. Overall, the value gained from investment in standard systems is subject tointernal uncertainties (with regards to system integration), external uncertainties (with regards tosocial demands and technology changes), and particularly interaction risk (with regards to networkeffects, evolving inter-agency collaborative capability and underlying social networks).

We have identified two obstacles to cross-agency standardization. First, standardization seems torequire individual agencies to give up the autonomies in managing their own systems to allow acentral commander to manage them on behalf of them. Second, many uncertain factors affect thevalues of standard systems, making investment in standard systems a highly risky business. Due tothese two barriers, investments in cross-agency standard systems and efforts to build collaborativeoperational capabilities have been minimally made in most nations.

To reduce the dilemma between empowering the central managers to assure the cross-agencyinteroperability and empowering the local managers to responsibly accomplish their uniquemissions, it is necessary to standardize common systems and interfaces, and to let individualagencies build specialized applications tailored to their own missions on top of these standardsystems. Malone (2004) notes that "Rigid standards in the right parts of a system can enable muchmore flexibility and decentralization in other parts of the system". Only those parts that are identifiedas involving important economies of scale without undermining the qualities of services need to bestandardized, and everything else can be decentralized to suit unique service needs.

The risks associated with investment in standard systems can be also mitigated by selective andadaptive standardization. The National Task Force on Interoperability (2003) emphasizes theimportance of strategic experimentation and adaptive implementation.

Improving interoperability is a complex endeavor. There are no "one size fits all"solutions. It may require agencies and jurisdictions to develop new and improvedworking relationships and could involve substantial changes in how individual agenciesoperate in terms of communication. Expect to make progress, but allow adequate timefor the progress to be substantial. Sometimes the most progress is made through smallsteps that test strategies and approaches. (National Task Force on Interoperability 2003)

Hypothesis 1:

Hypothesis 2:

1.7

1.8

1.9

1.10

steps that test strategies and approaches. (National Task Force on Interoperability 2003)

This research attempts to test two hypotheses.

Decentralized investments in interoperable and standard systems by autonomousagencies can substantially improve joint mission capabilities.

Although the values of standard systems are subject to various uncertainties, thevariances of outcomes generated from standard systems will not increase significantly ifinvestment in standard systems are made modestly and adaptively.

In sum, decentralized and adaptive IT investments along with modest standardization—withoutrequiring radical changes toward centralized IT management—can improve joint mission capabilitiessubstantially and robustly.

Kaplan and Norton (2001) list the difficulties of estimating the contribution of information systems tovalue creation: 1) Improvement in information systems does not directly affect final outcomes butonly through chains of cause-and-effect involving two or three intermediate stages. 2) Informationsystems have only potential value (i.e. system development cost is a poor approximation of anyrealizable value), and organization process are required to transform this potential value to realizedvalue. 3) The value of a system is interdependent with that of the systems with which it is networked.These difficulties are especially evident with standard systems that indirectly affect a wide range ofsystems and inter-organizational processes. Despite many manuals for IT investment, investmentsin improving interoperability and building standard systems have rarely benefited from the simulationof adaptive policy implementation.

This research attempts to validate the long-term dynamic effects of decentralized and adaptiveinvestments in interoperable and standard systems, using an agent-based modeling methodologythat simulates the uncontrolled dynamics that independent actors in a network may interactivelycreate. As shown in Table 1, the ultimate outcome of IT investment is 'joint mission capability', but notinteroperable system capacity that is just intermediate output. Although a production function for jointmission capabilities is given as an exogenously imposed system of equations, each agency isdesigned to autonomously and interactively determine its own IT portfolio, standardization policy, andsystem operation (no central commander controls information systems of individual agencies). Theproductivity of interoperable systems will be explicitly modeled as an endogenous social networkfunction instead of being left as exogenously given parameters. Exploratory modeling—by varying thehighly uncertain parameters for technology progress, interdependencies, and inter-agency systemutilization—will test the robustness of each policy alternative in terms of the variances of outcomes.

Table 1: Strategic Hypotheses of Key Measures, Policy Levers andUnderlying Uncertainty

UltimateOutcomes

Intermediate Outputs Long-termDrivers

Measures JointMissioncapabilities

Interoperablecapacities (informationcapital)Productivities (social capital)

Standardcapacities

Systemutilization rates

ExogenousUncertainty

Citizens' prioritiesof missioncapabilities

Interdependenciesamong systems formissions

Technologyadvance,Social networkdensity

Relationships Production function,System utilizationfunction

Technologydiffusion,Social networkdrivers

2.1

2.2

2.3

Policy Levers IT investment portfolioStandardization,Organizationalincentives

Assumptions for Network Economics

Overview and Relationships

This study implements an agent-based computational simulation using 'Blanche' software (version4.6.5)[1]. The objects that make up a model are nodes, attributes, and relations. A node represents agovernment agency, and an attribute is a numerical value that defines a property of a node. Eachnode has heterogeneous attributes. A relation is a set of numerical values that define interactionsamong N nodes using an N by N matrix. Each relation has an equation that describes how itsinteractions with other autonomous agents mutually and endogenously change over time. Forinstance, agencies often do not sufficiently invest in coupling complementary but highly disparatesystems of other agencies due to limited awareness and difficulty in integrating incongruent systems.Even after systems of other agencies become technically interoperable, the limited understandingabout how data are created and used may still constrain their utilization. Hence, the degree ofsimilarity between agencies can affect not only the investment in improving interoperability but alsothe readiness to utilize interoperable systems across agencies.

By referring to the Business Reference Model of the U.S. Federal Enterprise Architecture (OMB2005), this modeling exercise defines 20 agencies as an illustrative body of the federal government.Each of 20 agencies builds its own system to serve its distinct mission[2]. The system of the i-thagency will be hereafter called the i-th system, and the mission of the j-th agency will be hereaftercalled the j-th mission ({i ∈ N : 1 ≤ i ≤ 20} and {j∈N : 1 ≤ j ≤ 20}).

This study will model heterogeneous relations (including network benefits and coupling costs) amongsystems that have heterogeneous service attributes and levels. Network benefits and coupling costsamong 20 systems will be unevenly and asymmetrically distributed, and described using discreteand non-parametric 20 by 20 matrices.

2.4

2.5

Figure 1. Matrix of Interdependencies among Systems and Missions of 20 Agencies(The Department of Interior has identified complex interdependencies among multiple agencies and multiple

missions as the matrix of binary values. By referring to this matrix, the contributions of complementary systemsto missions are scored between 0 and 5, and the contributions of the own systems (diagonal cells) are scored

as 20. Then, these scores are normalized to the relative contribution of the i-th system to the j-th mission (wi.j ).)

Figure 1 illustrates the matrix of interdependent network benefits—i.e. the relative contributions of thei-th system to the j-th mission (denoted as wi.j). Coupling costs are derived from disparity betweensystems (denoted as di.q). The sophistication levels for the i-th system (denoted as Pi.q) areassessed along 11 service attributes (e.g. citizen services, e-Business, multimedia data, andtechnology level), and disparities between the i-th and the j-th system are calculated as: dij =

Σq=111(Max(Pi.q—Pj.q , 0)).

Network Values: Production Function of Mission Capabilities

We now introduce the concept of a core system which is defined as the unique system of an agencyto serve its own principal mission. The service level of the core system will be called 'core (system)capacity', and the core system capacity of the i-th agency will be denoted as xi . A core system can

2.6

2.7

2.8

2.9

2.10

also generate complementary values to other agencies serving different missions. However,heterogeneous core systems are incompatible with each other, so middleware or interfaces areneeded to make a core system interoperable with another core system. 'Interoperable capacity' of thei-th system for the j-th mission (denoted as xi,j i ≠ j ) is a portion of the core system capacity of the i-thagency that is interoperable with the core system of the j-th agency ( i ≠ j) and hence contributable tothe j-th mission (see Figure 3).

The mission capability (Uj) in this exercise is jointly produced from a core system and nineteeninteroperable systems that are operated through inter-agency interactions. A multiplicative (asopposed to additive) functional form is adopted to model the interdependent effects amongheterogeneous systems. The relative contributions of the i-th system to the j-th mission (denoted aswi.j)

are assumed to be exogenously given. This production function is assumed to exhibit constant

returns to scale (i.e. Σ i=120

wij = 1). The mission capability equation for the j-th mission, denoted asUj , is given by:

U j = Aj x1.j w1.j x2.j

w2.j x3.j w3.j … x20.j

w20.j (1)

Aj represents the total factor productivity, that is, the joint productivity of the twenty interdependentsystems. To define the total factor productivity, the concept of an 'enacted (or activated) interoperablecapacity' (denoted as exi,j) is introduced. Technically interoperable capacity (xij.t) can be transformedto enacted interoperable capacity through organizational efforts to utilize interoperable systemsacross agencies. That is, the gap between the current technically interoperable capacity (xij.t) andthe previous enacted interoperable capacity (exij.(t-1)) is filled by adding the utilization rate (denotedas φij ; 0 ≤ φij ≤ 1) times the gap at each time step. As a dynamic model, time step is indexed usingsubscript "t".

exij.t = φij.t (xij.t - exij.(t-1)) + exij.(t-1) = φij.t xij.t + (1 - φij.t) exij.(t-1) (2)

This utilization rate will be defined using social network parameters as shown in equation (10).Productivity of a given agency will be higher when the agency is more innovative (measured bytechnology level) and more actively utilizing interoperable capacities. Hence, this exercise definesproductivity as a function of technology levels and enacted interoperable capacities. The productivityequation for the j-th mission is given by (0 ≤ Aj ≤ 1):

Aj = w1.j (ex1.j / x 1.j ) + … + wj.j

(Techj / Techmax) + … + w20.j(ex20.j / x 20.j )

(3)

Cost Functions for Investment Alternatives

A total federal IT budget (denoted as M t) is allocated to core system capacities (M i.t), interoperablecapacities (M i.j.t i ≠ j ), mission-centric standard capacities (Mall.j.t) and government-wide standardcapacities (Mall.all.t). A cost function converts budgets (money) into system capacities. This exerciseassumes that system capacities incrementally expand as sequential modules over time (i.e. xi.j.t =

C-1(M i.j.t) + xi.j.t-1).

This exercise assumes that the cost of developing core system capacity is a convex function with aninverse of evolving technology level (which will be defined in equation 11).

(4)

2.11

2.12

2.13

2.14

3.1

3.2

Core systems of other agencies can be reused via bilateral point-to-point interfaces. Developing aninterface to couple and reuse an existing system is assumed to be scale-independent, so thecoupling cost is defined as a linear function of interoperable capacity and distance with an inverse ofevolving technology level.

(5)

In order to fully couple 20 systems with bilateral point-to-point interfaces, 380 (=20 × 19) interfacesare needed. Alternatively, mission-centric or government-wide standard systems may be built. Thecost of developing a standard system is assumed to be proportional to the needed scope to coverthe given heterogeneity. The needed standard scope for a mission-centric interoperability, Sj , isdefined as the sum of the distances between the j-th agency and all other agencies. The neededstandard scope for a government-wide common standard, Sall , is the sum of distances between themaximum (Pmax.q) and the minimum (Pmin.q) across 11 service categories.

(6)

(δ : the rate of cost saving due to standardization (0<δ<1)[3] )

(7)

While building a standard system is at least Si or Sall times as costly as building non-standardizedincompatible systems, doing so saves additional investments in numerous point-to-point interfacessince it is designed to be interoperable with all other systems. The mission-centric standard capacityis added to all of 19 interoperable capacities for a given mission (xi,all i ≠ j ), and the government-widestandard capacity is added to all of 380 interoperable capacities (xall,all i ≠ j ).

In sum, core system capacity and interoperable capacity build up over time as follows:

(8)

(9)

Agent-based Modeling of Network Dynamics

Network-centric public services are realized through investment in and operation of cross-agencyinteroperable capacities. Using an agent-based modeling framework that combines networkeconomics and social network analysis, this exercise will simulate not only the effects of inter-agency IT investment policy levers but also the effects of inter-agency system operation policy leversto better understand the co-evolution between technology and organization. Particularly, socialnetwork analysis methodology will be applied to model inter-agency collaborative operation, i.e. theuncontrolled dynamics that independent actors in a network may interactively create.

Social Network Drivers

Because the organizational mechanism of intergovernmental operations is complex and manyfactors influence it, this exercises adopts the multi-theoretical, multi-level social network modelingapproach (Monge and Noshir 2003). Various social network theories have attended to a wide range ofvariables such as social influence, power, diffusion, economic exchange, social cohesion, knowledge

3.3

3.4

3.5

3.6

3.7

3.8

management and social capital (Katz and Lazer 2002). Multiple theories can jointly improve anexplanation of network evolution. Multiple levels of analysis are also needed because networkproperties exist at the individual, dyad, triad, and global network levels (e.g. reciprocity at the dyadlevel and transitivity at the triad level).

In the absence of empirical data on inter-organizational operation of a standard system, this exercisewill simulate the hypothetical effects of six social network drivers (similarity, reciprocity, centrality,mission priority, interdependencies, and transitivity) to illustrate how autonomous agents operatethem in a decentralized environment over time.

In the case of the first characteristic, social networks can be biased toward similarity (e.g. technicaland semantic similarity) since similarity may ease communication and increase the predictability ofbehavior. In the case of the second, reciprocity often plays an important role in building trust in non-hierarchical relationships. In the case of the third, centrality in relations is important. Barabasi andBonabeau (2003) assert that preferential attachment (or the "rich get richer" process) explains thegrowth of hubs. That is, new nodes tend to connect to popular nodes, so the hubs acquire even morelinks over time than the less connected nodes. In the case of the fourth, agencies may tend to buildmore social links with other agencies in charge of more socially demanded missions. In the case ofthe fifth, an agency may build more social links with other agencies that operate highlycomplementary systems. Finally, an agency may develop relationships with another agency throughother (intermediary) agencies that have close ties with that agency. This indirect relationship is called'transitivity', and this may help extend social networks with other agencies.

This exercise defines the utilization rate (denoted as φij) of the interoperable capacity xi,j as afunction of these six social network drivers between the i-th agency and the j-th agency:

φij = (φ1.ijκ1 φ2.ij

κ2 φ3.ijκ3 φ4.ij

κ4 φ5.ijκ5 φ6.ij

κ6 )τ(0 ≤ φij ≤ 1) (10)

(τ: scale factor; κk : weight of the k -th factor)

These six factors are powered by weights (denoted as κ1~κ6) and multiplied together to constructthe inter-agency system utilization rate function. Inter-agency system operation policy is assumed toaffect these weights to six factors. The multiplied value of the six factors is between 0 and 1, so whenthe scale factor (denoted as τ) in the exponent is less than one, the utilization ratio becomesmagnified.

Technology Innovations and Diffusions

The technology equation consists of an innovation equation and an imitation equation in a continuousscale. Technology level for the i-th agency (Techi.t) is defined as:

Techi.t = TechInovi.t-1 × Techgr i + Σ j=120((exji / Σ ex ji) × φij ×

Max(Techj(t-1) -TechInovi.t , 0))(11)

The technology innovation equation (denoted as TechInovi.t) assumes that agencies build uptechnology level with heterogeneous technology growth rates (denoted as Techgri ; 1.05 ≤ Techgri ≤1.105). While agencies with higher technology growth rates build up their technology levels on theirown, other agencies catch up with them through imitation. Social network plays an important role intechnology diffusion, so the technology imitation equation includes the utilization rate φij . Thetechnology imitation function is the weighted average of the technology gaps multiplied by theutilization rates (weighted by exji / Σ j=1

20exji). While different innovation capacities widentechnological distance over time, technology imitations reduce technological distance.

Defining Policy Alternatives

4.1

4.2

4.3

4.4

This modeling exercise aims to simulate the joint effects of three policy levers (IT investmentportfolio, standardization, and inter-agency operation). For IT investment portfolio lever and inter-agency operation policy lever, this study will compare 'agency-centric (or supplier-centric)' and'mission-centric (or citizen-centric)' approaches.

Agency-centric vs. Mission-centric IT investment portfolio: Individual agencies allocate a givenbudget to the development of not only their own systems but also interfaces and middlewarethat couple complementary systems of other agencies. The investment portfolio is often biasedtoward their own systems and linkages with similar systems, and this study will call such abiased portfolio 'agency-centric'. When the investment portfolio is aligned with the optimalinterdependencies without bias, such portfolio will be called 'mission-centric'.Agency-centric vs. Mission-centric operation of interoperable systems: This study assumesthat six social network drivers jointly determine inter-agency system operation. When similarityand reciprocity are dominant drivers, this study will call such operations 'agency-centric'.When cross-agency interdependencies and transitivity are dominant drivers, such operationswill be called 'mission-centric'.Bilateral coupling vs. Multi-lateral standardization: With bilateral coupling alternatives,complementary systems are integrated via point-to-point interfaces. Standardization can bemade for specific missions (e.g. public safety, and healthcare) or for government-wide sharedservices and protocols (e.g. next-generation Internet protocol, authentication, and informationsecurity). Standardizing government-wide shared services requires a high level of cross-agency coordination.

This exercise derives five alternatives (Alt0 - Alt4) as joint implementation of three policy levers. Alt0 isa business-as-usual ('do-nothing') policy that maintains 'agency-centric IT investment portfolio withno investment in standard system and agency-centric operation of interoperable systems'. Since itprovides a baseline for evaluation of the joint effects of three policy levers, it will be called 'baselinepolicy'. Alt1, Alt2 and Alt3 are 'do-something' alternatives that implement one or two policy levers. Alt4is a comprehensive ('do-everything') policy that implements mission-centric IT portfolio and operationwith modest investment in not only mission-centric standard systems but also government-widestandard systems.

All the five policy alternatives are implemented using the same amount of federal IT budget (M t), and

the federal budget is allocated to the j-th agency according to its mission weight (i.e. M j.t = wj.t M t)[4].

Decision variables for each agency to implement three policy levers are: 1) M i.j.t, 2) Mall.j.t, Mall.t and3) κ1~κ6 in φij .

The baseline policy (Alt0) allocates the budgets proportionally to each system's contribution weightsand the inverse of distance (M i.j.t ∝ wi.j.t / dijt). On the other hand, the other alternatives implementthe mission-centric IT portfolio policy by allocating the budgets proportionally to the system'scontribution weights (M i.j.t ∝ wi.j.t)

[5]. On top of the mission-centric IT portfolio policy lever, fouralternatives (Alt1-4) are generated as the combinations of the two longer-term drivers:standardization policy and inter-agency operational policy (see Table 2).

Table 2: Characteristics of Four Alternative Mission-centric IT InvestmentPortfolio Policies

OperationIntegration Within Boundary

(reinforcingfragmentation)

Across Boundary(promoting cross-agency

collaboration)No Standardization Alternative 1 :

• No standard system Alternative 3 :• No standard system

4.5

4.6

• Inflexible budgetreallocation• Homophily,Reciprocity, Inequality inrelations

• Flexible budgetreallocation• Promoting socialnetwork and demand-driven system operation

ModestStandardization(reallocates modestportion of budgets tothe development ofstandard systems)

Alternative 2 :• Mission-centricstandard systems• Inflexible budgetreallocation• Homophily,Reciprocity, Inequality inrelations

Alternative 4 :• Government-widestandard systems• Flexible budgetreallocation• Promoting socialnetwork and demand-driven system operation

Standardization policies (Alt2 and Alt4) allocate a small portion (five percent[6]) of IT budgets to thedevelopment of standard systems. While Alt2 (agency-centric operation) builds only decentralizedmission-centric standard systems (Mall.j.t=.05 M j.t), Alt4 (mission-centric operation) is assumed tobuild coordinated government-wide standard systems (Mall.t=.02 M t) as well as decentralizedmission-centric standard systems (Mall.j.t=.03 M j.t).

The inter-agency operational policy lever is simulated using two sets of weights assigned to the sixsocial network drivers in the system utilization equation (κ1~κ6 in φij) as shown in Table 3.

Table 3: Social Network Drivers and their Contribution Weights toUtilization Rates

Social Network Drivers κ1~κ6 in φij

Agency-centricsystem operation

(Alt0,1&2)

Mission-centricsystem operation

(Alt3&4)Relative Distance between the i-thand j-th agency (=dmin/dij)

0.3 0.1

Relative Reciprocity in relationswith j-th agency (=exji/exmax)

0.2 0.1

Relative Degree Centrality* of j-thagency (=centj/centmax)

0.2 0.2

Relative Priority of the j-th mission(=wj/wmax)

0.1 0.1

Relative Contribution of the i-thsystem xij (=wij/wmax)

0.1 0.3

Relative Transitivity in relations (i-k-j)**

0.1 0.2

* Degree centrality measures the number of incoming and/or outgoing connections with others.** The transitive relations of the i-th agency with j-th agency via all other (intermediary k -th) nodesare computed as tr ij = .05 Σk =1

20 (exik + exkj)/2. They are averaged across k (weighted by thevalue of links with intermediary node k ). Relative transitive relations are computed as (tr ij / ex ij).

Results of Agent-based Modeling of Network Dynamics

5.1 This exercise adopts a cost-effectiveness analysis framework that searches for the maximumeffectiveness subject to a given budget. The key measure of effectiveness is joint mission capabilityat the final period. Each mission capability (Uj) is defined as an independent building-block, and the

joint value of mission capabilities is defined as the weighted average: Σ j20

wjUj . The mission

weights (i.e. the relative importance of mission j; denoted as wj) are assumed to be exogenous[7] and

normalized ( Σ j20

wj = 1). Each time step represents one year, and this exercise runs for 20 timesteps to simulate long-term effects. Figure 2 shows the evolution of the average system capacities,productivities, and mission capabilities generated from this simulation. The mission-centric ITportfolio with mission-centric standards (Alt2) builds much larger interoperable capacities thanbilateral coupling policies (Alt0,1&3), and the mission-centric IT portfolio including both mission-centric and government-wide standards (Alt4) achieves by far the largest interoperable capacities.Productivities rapidly drop with the upsurge of interoperable capacities generated fromstandardization (Alt2&4) in the early periods, but productivity soon rebounds with the mission-centricoperation policy (Alt4).

5.2

5.3

Figure 2. Evolution of System Capacity, Productivity, and Mission Capability

A remarkable enhancement of the joint mission capabilities beyond the baseline level (Alt0) comesfrom the adoption of mission-centric IT portfolio policy (Alt1) through its better alignment with theoptimally diversified system portfolio. Once the mission-centric IT investment portfolio policy isimplemented, either standardization (Alt2) or mission-centric operation (Alt3) increases the missioncapability further. When investment in standard systems and mission-centric operation aresimultaneously implemented (Alt4), their joint effects on joint mission capabilities are substantiallyimproved through positive feedback loops between interoperable capacities and productivities overtime. The gain of alternative 2 over alternative 1 is 26%, the gain of alternative 3 over alternative 1 is20%, and the gain of alternative 4 over alternative 1 is 71% (far exceeding the sum of the gains fromalternative 2 and 3). Nonetheless, this gain of mission capabilities (71%) is not as dramatic as thegain of interoperable system capacities (152%) due to the untargeted expansion of interoperablecapacities by standardization.

The graphs and matrices in Figure 3 present the technically interoperable capacities (xi,j) and

6.1

6.2

enacted (activated) interoperable capacities (exi,j) generated from the simulation exercise. While thesum of activated interoperable capacities (1340) by the DOD with all other agencies is just four timesits core system capacity (381), the sum of activated interoperable capacities (484) by the GSA farexceeds ten times its core system capacity (36). This shows that agencies in charge of government-wide common management and support functions can create more values from expanding andactivating inter-agency interoperable systems rather than from building its own core systemcapacities.

Figure 3. Interoperable System Network and Activated System Network (Alternative 4 attime=20; presented using the UCINET software)

In the above matrix and graph, the diagonal cells and the node sizes represent the core system capacities, andthe non-diagonal cells and the thicknesses of arcs represent the interoperable capacities. Likewise, in the below

matrix and graph, the diagonal cells and the node sizes represent the degree centrality (i.e. the sum of all thelinks) of each agency, and the non-diagonal cells and the thicknesses of arcs represent the activated(enacted)

interoperable capacities.

Exploratory Modeling for Assessing the Robustness

Identifying Critical Uncertainties and Generating Scenarios

Some of parameter values adopted in the previous simulation exercise are in fact highly uncertain.The priorities citizens assign to various missions may change as economic and social environmentsevolve. Technology progress is hard to predict precisely. The interdependencies amongcomplementary systems for mission capabilities also continue to change. The inter-governmentalrelations such as technology diffusion and social networking among federal agencies are verycomplex. All these uncertain parameters jointly affect the robustness of final outcomes.

Since policy-makers and citizens are often risk-averse, reducing the sensitivity of outcomes againstvarying parameters are as important as increasing the levels of the most likely outcomes for eachalternative. This exercise hypothesizes that technology development, mission priority, system-mission interdependency, and system utilization rate are the most uncertain factors.

6.3

6.4

6.5

6.6

The baseline parameters values used in the previous section are: the scale factor (τ) in the utilizationrate equation = 1, the average technology growth rate = 7.5%, and the increasing weights tocomplementary systems (wi,j i ≠ j )

[8]. By taking advantage of both parametric and probabilisticexploratory modeling in a complementary way (as shown in Table 4), this study will test therobustness of policy alternatives. Parametric exploratory analysis will generate 18 scenarios as thecombination of multiple values for three uncertain parameters. Probabilistic exploratory analysis willgenerate 100 simulations as the combination of parameter values randomly drawn from theprobabilistic distribution functions.

Table 4: Values for Uncertain Parameters

Parametricmodeling

Probabilisticmodeling

Weights of the i-th systems for thej-th mission (wi,j i ≠ j )

Constant, andIncreasing (1%annual growth)

N(1.01, 0.03):normal distribution

Technology growth (Techgr i) 4%, 7.5% and 11%on average

Poissondistribution*(μ = σ =7.5%)

The scale factor (τ) of theutilization rate function**

0.5, 1, and 1.5 N(1, 0.05)

Weights of missions (wj) N(1, 0.03)

* Given the lock-in effects arising from network standards, information technologies are more likelydisruptively leaping than incrementally improving. Hence, the probabilistic modeling of technologyinnovation equation will adopt a Poisson distribution function.** Since the utilization rate lies between 0 and 1, a higher exponent means a lower utilizationscenario. High scale factor of the utilization rate function may mean that social capital acrossagencies is high.

Parametric Exploratory Modeling

Figure 4 shows the weighed average of mission capabilities. The robustness of each alternative canbe measured in terms of the coefficients of variation (= standard deviation / mean). The coefficients ofvariation across 18 scenarios for five alternatives are : 28.3% (Alt0), 33.6% (Alt1), 29.9% (Alt2),30.7% (Alt3), and 25.6% (Alt4). The outcomes of most mission-centric IT investment portfolioalternatives (Alt1-Alt3) turn out to be more sensitive to various uncertainty factors than the baselinescenario. However, the comprehensive policy including modest investment in government-widestandard systems (Alt4) minimizes the coefficient of variation while maximizing the expected values.

Probabilistic Exploratory Modeling

The probabilistic exploratory modeling has generated stochastically-evolving mission priority weights(wj) and system's relative contribution weights (wi.j) over time. To respond to changing missionweights over time, budget allocation has been adaptively made proportional to the magnitude of gapsbetween the desirable portfolio and the current portfolio.

Standardization policies (Alt2&4) mitigate the coefficients of variation of the interoperable capacities,and the mission-centric operation policies (Alt3&4) mitigate the coefficients of variation of theproductivities. Again, as shown in Figure 5, the comprehensive policy including modest investment ingovernment-wide standard systems (Alt4) minimizes the coefficient of variation of the missioncapabilities while maximizing their expected outcome.

7.1

7.2

Figure 4. Joint Mission Capabilities from the Parametric Exploratory Modeling

Figure 5. Joint Mission Capabilities Generated from Probabilistic Exploratory Modeling

Summary of Modeling and Policy Implications

While these modeling exercises are highly abstract, speculative and merely suggestive, the resultsdemonstrate that when autonomous agencies build more mission-centric (rather than agency-centric) IT portfolios, invest in standard systems modestly, and build more mission-centric (ratherthan agency-centric) relationships with other agencies, an encouraging evolutionary trajectory can beproduced without any central commander.

Standardization improves interoperability among a broad range of systems while point-to-pointinterfaces selectively couple only highly complementary systems. Building operational capabilities

7.3

7.4

that activate a wide range of interoperable capacities takes time. Hence, small investments instandard systems improve a wide range of interoperability remarkably (by 152%), but suchuntargeted interoperable capacities with lagging operational capabilities improve joint missioncapability less remarkably (by 71%). Nonetheless, such untargeted interoperable capacities mayimprove the robustness of mission capabilities faced with highly uncertain future states. Both theparametric and the probabilistic exploratory modeling confirm that modest investment in standardsystems jointly with mission-centric operation not only enhances the expected outcome but alsoreduces the variances of outcomes against varying parameters for technology progress,interdependency, and inter-agency operation.

These findings confirm our two hypotheses, demonstrating that decentralized and adaptiveinvestments in interoperable and standard systems—as long as individual agencies adopt mission-centric IT investment and operation approaches—can improve joint mission capabilities substantiallyand robustly without requiring radical changes toward centralized IT management.

Identifying appropriate scale or scope of standard system remains to be a difficult task for furtherresearch. This simulation exercise illustrates that a number of cross-agency social network driverscan significantly and substantially affect inter-agency operational readiness, and consequently theadded values of standard systems. Empirical research on inter-agency social network mechanismscan help better estimate the productivities of cross-agency interoperable and standard systems, andhence provide useful information for deciding appropriate scale and scope of cross-agency standardsystems.

Notes

1Blanche is a program designed to evaluate a hypothesis by simulating computational models ofevolutionary network dynamics. Blanche is developed under the direction of N. Contractor at theUniversity of Illinois at Urbana-Champaign.

2The twenty distinct missions include 16 citizen service missions (Community and social services,Defense, Economic development, Education, Energy, Environmental management, General science,Health, Homeland security, Income security, International affairs, Law enforcement, Naturalresources, Transportation, Workforce management, and Revenue collection and Finance) and fourmanagement functions (HR management, Public asset management, Public records management,and Budget planning).

3In this simulation, δ is assumed to be 0.3.

4The initial value of the total IT budget (M0) and mission weights (wj) are derived from the actual U.S.IT budget in FY2003. This simulation assumes 7 percent annual growth of the federal IT budget(M t=1.07t-1 M0).

5If the budget for a point-to-point interface is allocated more than needed to exploit the full capacity ofthe original system, the agency is assumed to either expend the surplus budget within itsorganizational boundary through building its own non-core system redundantly (the agency-centricsystem operation policies: Alt0,1&2) or transfer the surplus budgets to the agency in charge of theoriginal system to consolidate the system development (the mission-centric system operationpolicies: Alt3&4).

6The 'e-Government fund' authorized during the Bush administration to support the Presidential e-Government initiatives is an example of investment in standard systems. The E-Gov fund ($345million over four years) is less than 0.2 percent of the federal IT budget. Five percent seems to be anambitious yet realistic target.

7The initial mission weights (at t=0) are assumed to be proportional to the actual federal IT budgets inFY 2003. 8The weights to the complementary systems grow by 1 percent each time step (wi,j.t =1.01 wi,j.t-1 ; i ≠ j ), and then all the weights are normalized.

Consequently, the average weight to own core systems (wi.i.t) gradually diminishes from 0.484 attime=1 to 0.435 at t=20.

References

AGRANOFF R (2003) Leveraging Networks: A Guide for Public Managers Working acrossOrganizations. IBM Endowment for the Business of Government.

BARABASI A and Bonabeau E (2003) Scale-free Networks. Scientific American, May, pp. 60-69.

FOUNTAIN J E (2001) Building the Virtual State: Information Technology and Institutional Challenge .Washington, D.C.: Brookings Institution Press.

FOUNTAIN J E (2002) Information, Institution and Governance. National Center for DigitalGovernment, JFK School of Government.

GOLDSMITH S and Eggers W D (2004) Governing by Network: The New Shape of the PublicSector. Washington D.C.: Brookings Institution Press.

KAPLAN R S and Norton D P (2001) The Strategy-Focused Organization: How Balanced ScorecardCompanies Thrive in the New Business Environment. Harvard Business School Press.

KATZ N and Lazer D (2002) Building effective intra-organizational networks: The role of teams .http://www.ksg.harvard.edu/teamwork/

KAYE D (2003) Loosely Coupled: The Missing Pieces of Web Services . California: RDS Press.

KOHTAMAKI M, Vuorinen T, Varamaki E, and Vesalainen J (2008) Analysing partnerships and strategicnetwork governance. International. Journal of Networking and Virtual Organisations , 5 (2) pp. 135—154.

LAZER D and Binz-Scharf M C (2004) Information Sharing in E-Government Projects: ManagingNovelty and Cross-Agency Cooperation. Report for IBM Endowment for the Business ofGovernment.

LIBICKI M (2000) Who Runs What in the Global Information Grid . RAND MR-1247-AE.

MALONE T W (2004) The Future of Work . Boston: Harvard Business School Press.

MONGE P R and Noshir S C (2003) Theories of Communication Networks . New York: OxfordUniversity Press.

NATIONAL TASK FORCE ON INTEROPERABILITY (2003) Why Can't We Talk?: Working Togetherto Bridge the Communication Gap to Save Lives.

OMB (Office of Management and Budget) (2005) FY07 Budget Formulation FEA ConsolidatedReference Model Document.